An Evaluation of Hybrid Deep Learning Models for Classifying Multiple Lower Limb Actions

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Brain-computer interfaces (BCIs) translate electroencephalography (EEG) signals into control commands.
• Motor imagery (MI) is a common BCI paradigm, often utilizing deep learning models like CNNs and LSTMs.
• Existing research has limitations in classifying separate lower limb movements.

Purpose of the Study:

• To explore and evaluate hybrid deep learning models (CNN-LSTM combinations) for classifying individual lower limb actions.
• To assess the suitability of these models for differentiating between motor imagery, real movements, and movement observations.
• To identify the most effective hybrid model for lower limb action classification in BCIs.

Main Methods:

• Implemented and compared four hybrid deep learning models combining Convolutional Neural Networks (CNNs) and Long Short-Term Memory (LSTM) networks.
• Utilized EEG data to classify motor imagery, real movements, and movement observations for individual lower limbs.
• Evaluated model performance based on classification accuracy.

Main Results:

• No single hybrid model demonstrated significantly superior classification accuracy over others.
• All tested hybrid models outperformed chance level performance in classifying lower limb actions.
• The study successfully classified multiple actions including motor imagery, real movements, and movement observations.

Conclusions:

• Hybrid CNN-LSTM models show potential for classifying separate lower limb actions in BCIs.
• The classification of multiple actions (MI, real, observed movements) is feasible with current hybrid models.
• Further research is needed to optimize models for individual lower limb action classification in BCI systems.